Paper Title
Development of an Anatomically-Awareensemble Deep Learning Framework for Automated Detection and Classification of Postural Kyphosis Using Sagittal Radiographs

Abstract
Postural kyphosis, a common spinal deformity marked by excessive thoracic curvature, is increasingly prevalent due to modern sedentary lifestyles, contributing to chronic pain, joint degeneration, and reduced quality of life. While awareness of its impact has grown, current diagnostic and intervention methods often lack scalability and real-world applicability. This paper addresses the need for advanced, automated approaches by proposing a novel anatomically-aware ensemble deep learning model for the detection and classification of postural kyphosis using sagittal radiographs. The proposed model incorporates anatomical feature localization, a diverse ensemble of deep learning backbones, and a dual-task head to simultaneously classify kyphosis severity and predict Cobb angle measurements. This research may demonstrate improved performance compared to standard CNN models, highlighting the framework’s potential to assist orthopedists and radiologists, improve diagnostic accuracy, and support timely, AI-driven intervention strategies. Keywords - Postural deviations, Postural Kyphosis, Chronic pain, Cobb angle, CNN model.